Originally published in Forbes Technology Council
As AI agents take on more operational work, from coordinating resources to triggering transactions, much of the conversation about risk has focused on security: keeping agents from being manipulated or given more access than they need. But what about an agent that’s secure, authorized and following its instructions, and still gets it wrong?
In his latest Forbes Technology Council article, Arango CEO Shekhar Iyer explores that risk: agents that act without the current, connected context they need to understand the consequences of their actions.
He offers a hurricane scenario. An AI agent coordinating an emergency response redirects supplies without knowing that a bridge has closed or that another team has already reserved the remaining vehicles. The agent did exactly what it was designed to do, using the only information it could see.
From Access to Context
Shekhar points to three common reasons agents miss critical context. They see records but not the relationships between them. Their information may be outdated, or its source can’t be verified. And when separate agents each assemble their own view, they can act on different versions of the same reality.
Before expanding an agent’s autonomy, he suggests leaders ask what could change its decision, how current that context must be and whether its evidence can be traced.
Key Takeaways
- An agent can stay within its guardrails and permissions and still act on incomplete or outdated information.
- Agents need context: relationships, current state, history, policies and trusted sources.
- Before giving an agent more authority, make sure it works from current, connected and governed context.
